app.py

import gradio as gr
import numpy as np
import random
import spaces
import torch
from diffusers import  DiffusionPipeline, FlowMatchEulerDiscreteScheduler, AutoencoderTiny, AutoencoderKL
from transformers import CLIPTextModel, CLIPTokenizer,T5EncoderModel, T5TokenizerFast
from live_preview_helpers import calculate_shift, retrieve_timesteps, flux_pipe_call_that_returns_an_iterable_of_images

from llm_wrapper import run_gemini
from huggingface_hub import hf_hub_download
from safetensors.torch import load_file
import subprocess


subprocess.run("rm -rf /data-nvme/zerogpu-offload/*", env={}, shell=True)


dtype = torch.bfloat16
device = "cuda" if torch.cuda.is_available() else "cpu"

taef1 = AutoencoderTiny.from_pretrained("madebyollin/taef1", torch_dtype=dtype).to(device)
good_vae = AutoencoderKL.from_pretrained("black-forest-labs/FLUX.1-dev", subfolder="vae", torch_dtype=dtype).to(device)
pipe = DiffusionPipeline.from_pretrained("black-forest-labs/FLUX.1-dev", torch_dtype=dtype, vae=taef1).to(device)

# PONIX mode load
pipe.load_lora_weights('cwhuh/ponix-generator-v0.1.0', weight_name='pytorch_lora_weights.safetensors')
embedding_path = hf_hub_download(repo_id='cwhuh/ponix-generator-v0.1.0', filename='./ponix-generator-v0.1.0_emb.safetensors', repo_type="model")
state_dict = load_file(embedding_path)
pipe.load_textual_inversion(state_dict["clip_l"], token=["<s0>", "<s1>", "<s2>"], text_encoder=pipe.text_encoder, tokenizer=pipe.tokenizer)

torch.cuda.empty_cache()

MAX_SEED = np.iinfo(np.int32).max
MAX_IMAGE_SIZE = 2048

pipe.flux_pipe_call_that_returns_an_iterable_of_images = flux_pipe_call_that_returns_an_iterable_of_images.__get__(pipe)

@spaces.GPU(duration=75)
def infer(prompt, seed=42, randomize_seed=False, width=1024, height=1024, guidance_scale=3.5, num_inference_steps=28, progress=gr.Progress(track_tqdm=True)):
    if randomize_seed:
        seed = random.randint(0, MAX_SEED)
    generator = torch.Generator().manual_seed(seed)

    if refine_prompt:
        refined_prompt = run_gemini(
            target_prompt=prompt,
            prompt_in_path="prompt.json",
        )
        print(f"Refined prompt: {refined_prompt}")
    else:
        refined_prompt = prompt
    
    for img in pipe.flux_pipe_call_that_returns_an_iterable_of_images(
            prompt=refined_prompt,
            guidance_scale=guidance_scale,
            num_inference_steps=num_inference_steps,
            width=width,
            height=height,
            generator=generator,
            output_type="pil",
            good_vae=good_vae,
        ):
            yield img, seed
    
examples = [
    "기계공학과(로켓) 포닉스",
    "바이올린을 연주하는 포닉스",
    "물리학을 연구하는 포닉스",
]

css="""
#col-container {
    margin: 0 auto;
    max-width: 520px;
}
"""

with gr.Blocks(css=css) as demo:
    
    with gr.Column(elem_id="col-container"):
        gr.Markdown(f"""# [POSTECH] PONIX Generator 🌊
([based on FLUX.1-dev](https://huggingface.co/black-forest-labs/FLUX.1-dev/blob/main/LICENSE.md))
        """)
        
        with gr.Row():
            
            prompt = gr.Text(
                label="Prompt",
                show_label=False,
                max_lines=1,
                placeholder="Enter your prompt",
                container=False,
            )  
            
            run_button = gr.Button("Run", scale=0)
        
        result = gr.Image(label="Result", show_label=False)
        
        with gr.Accordion("Advanced Settings", open=False):

            with gr.Row():

                refine_prompt = gr.Checkbox(label="Refine prompt", value=True)
                
                seed = gr.Slider(
                    label="Seed",
                    minimum=0,
                    maximum=MAX_SEED,
                    step=1,
                    value=0,
                )
                
                randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
            
            with gr.Row():
                
                width = gr.Slider(
                    label="Width",
                    minimum=256,
                    maximum=MAX_IMAGE_SIZE,
                    step=32,
                    value=1024,
                )
                
                height = gr.Slider(
                    label="Height",
                    minimum=256,
                    maximum=MAX_IMAGE_SIZE,
                    step=32,
                    value=1024,
                )
            
            with gr.Row():

                guidance_scale = gr.Slider(
                    label="Guidance Scale",
                    minimum=1,
                    maximum=15,
                    step=0.1,
                    value=3.5,
                )
  
                num_inference_steps = gr.Slider(
                    label="Number of inference steps",
                    minimum=1,
                    maximum=50,
                    step=1,
                    value=28,
                )
        
        gr.Examples(
            examples = examples,
            fn = infer,
            inputs = [prompt],
            outputs = [result, seed],
            cache_examples="lazy"
        )

    gr.on(
        triggers=[run_button.click, prompt.submit],
        fn = infer,
        inputs = [prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps],
        outputs = [result, seed]
    )

demo.launch()